PcMark: Note 3:Lollipop vs Note 4

Search This thread

fixfox2

Senior Member
Aug 9, 2010
247
68
Real world use rarely max's the cpu so this attached score means very little.
My daily use of the phones however suggests that note4 is poorly optimised (lollipop ). ....if indeed it is faster in hardware.
 

Attachments

  • 1428704122568.jpg
    1428704122568.jpg
    59.8 KB · Views: 84

v1rtu4l

Senior Member
Jul 18, 2008
323
98
You mean because the clock speed is rarely maxed? That's not really a big issue to be honest. A benchmark that fully loaded the CPU wouldn't tell you a lot because the device would be throttling within 10 seconds so wouldn't give you any indication of real usage at all.

no,
my original thought was if the difference in benchmark scores between the note 3 exynos and the note 4 snapdragon is due to the cpu or the optimization of it, then at least it should show a bit of load on the cpu while performing the tasks.

the screenshot of blackbahamuts benchmark result (of his exynos note 4) is very telling. even on stock firmware he has many CPU spikes in certain parts of the benchmark. he has more than 15 cpu spikes some even for a longer period of time which indicated the cpu being under full load meaning the cpu is not fast enough to process all instructions and so it needs all the cpu clock speed to process it.

these spikes are not present in the snapdragon version. i have exactly one short cpu spike for the whole benchmark (without power saving activated) to put this into perspective.
so the exynos processor seems to be a bottleneck here actually.


this leads me to the conclusion that it is probably only the cpu gouvernor that is differently tweaked and much more lazy on the snapdragon, so the CPU will not even raise the speed up for this type of task. i am pretty sure that if you modify the cpu gouvernor and set it to raise the CPU speed instantly it will show similar results. since the benchmark is probably not even multi-threaded (since there is always only one task being done) the amount of cores should be irrelevant and in the end theoretically only the raw cpu clock speed should count, since exynos and snapdragon both use arm architecture with very similar if instruction sets.

back in the old days on the galaxy s2 i had a mod which allowed me to change the cpu governor settings on the fly. if there is such a mod already for the note 4 snapdragon it would be nice to let somebody run the benchmark with stock settings and then with a governor setting tweaked for full performance.

on the other hand it is also reasonable to assume that the benchmark only really makes use of less cpu power than is available on the snapdragon variant.

snapdragon variant has a top speed of 2.7 GHz and hits only one spike during the benchmark.
exynos has a top speed of 1.9 GHz and hits the 100% quite a lot. so the benchmark is probably giving the cpu hardly enough to do fill the pipe for a 2 GHz cpu.

all these observations just lead to the cpu not really being a big factor in the benchmark. in the end most of the time the storage speed is more important anyways, because todays RAM and CPUs are so fast you that you got trouble to get the data read from flash storage to the RAM in the first place and that is what delays the operation and not the cpu not having the power to process all the instructions you throw at it.

just had a look at the galaxy note 3 cpu graphs and they do not even spike much.
the part where the note 3 destroys the note 4 snapdragon is web browsing (probably better optimized web rendering engine) and writing. it is pretty probable that the note 3 just has a faster internal storage or some other factor that makes it write to internal storage faster than the note 4.
 
Last edited:

darkbahamut

Senior Member
Mar 1, 2010
126
48
no,
my original thought was if the difference in benchmark scores between the note 3 exynos and the note 4 snapdragon is due to the cpu or the optimization of it, then at least it should show a bit of load on the cpu while performing the tasks.

the screenshot of blackbahamuts benchmark result (of his exynos note 4) is very telling. even on stock firmware he has many CPU spikes in certain parts of the benchmark. he has more than 15 cpu spikes some even for a longer period of time which indicated the cpu being under full load meaning the cpu is not fast enough to process all instructions and so it needs all the cpu clock speed to process it.

these spikes are not present in the snapdragon version. i have exactly one short cpu spike for the whole benchmark (without power saving activated) to put this into perspective.
so the exynos processor seems to be a bottleneck here actually.


this leads me to the conclusion that it is probably only the cpu gouvernor that is differently tweaked and much more lazy on the snapdragon, so the CPU will not even raise the speed up for this type of task. i am pretty sure that if you modify the cpu gouvernor and set it to raise the CPU speed instantly it will show similar results. since the benchmark is probably not even multi-threaded (since there is always only one task being done) the amount of cores should be irrelevant and in the end theoretically only the raw cpu clock speed should count, since exynos and snapdragon both use arm architecture with very similar if instruction sets.

back in the old days on the galaxy s2 i had a mod which allowed me to change the cpu governor settings on the fly. if there is such a mod already for the note 4 snapdragon it would be nice to let somebody run the benchmark with stock settings and then with a governor setting tweaked for full performance.

on the other hand it is also reasonable to assume that the benchmark only really makes use of less cpu power than is available on the snapdragon variant.

snapdragon variant has a top speed of 2.7 GHz and hits only one spike during the benchmark.
exynos has a top speed of 1.9 GHz and hits the 100% quite a lot. so the benchmark is probably giving the cpu hardly enough to do fill the pipe for a 2 GHz cpu.

all these observations just lead to the cpu not really being a big factor in the benchmark. in the end most of the time the storage speed is more important anyways, because todays RAM and CPUs are so fast you that you got trouble to get the data read from flash storage to the RAM in the first place and that is what delays the operation and not the cpu not having the power to process all the instructions you throw at it.

would be interesting to the see cpu graph for the note 3 versions actually. do they spike a lot ?

Some of the spiking is due to big.LITTLE as the graph is spiking between full speed on the A57's and full speed on the A53's and sometimes lower speeds on the A53's. Changing clock speeds is a fast operation so unless it is actually lowering performance then it isn't much of an issue (or bottleneck). It's more important to try and avoid unnecessary high speeds on power hungry cores like the A57's so it may just be that the governor is more aggressive at down clocking once the load drops.

I'll try locking my cores to stop them downclocking and see if it makes any difference.

Rerun of normal speed range: 5556 - http://i101.photobucket.com/albums/m42/emeraldphoenix/2015-04-10 23.47.46_zpslyiqebb7.png~original

Speed lock: 5769 - http://i101.photobucket.com/albums/m42/emeraldphoenix/2015-04-10 23.54.01_zps1vr8bjqm.png~original

2.5% speed improvement. Clearly the cores running down at 700Mhz at times isn't really reducing performance very much, but it will save a lot of power in the real world.
 

v1rtu4l

Senior Member
Jul 18, 2008
323
98
that is pretty much what i expected.
the end result will not change much because the difference in scores is in the parts that are not so cpu intensive like web browsing and writing, but the little parts of the benchmark where cpu spikes happen will benefit from a forced high clock speed.

you can even tell that only the parts of the benchmark (web & write) benefitted from forced high clock speed that had spikes in the first place. the other parts of the benchmark did not improve at all.

of course this is just virtual e-peen tweaking for useless benchmark results, so it is not sensible to have such settings in real life (at least not of you plan to plug your phone off the power cable :p )

i think it is save to say that the cpu does not play a big part in the bench mark at all since the note 3 is already a powerfull beast cpu-wise. this also kicks out the assumption that samsung did not properly optimize for the snapdragon, if the cpu does not play a big part in the results in the first place.
 
Last edited:

darkbahamut

Senior Member
Mar 1, 2010
126
48
Actually, if you look at the breakdown you can see the web browsing score increased by 15% and all the other tests stayed the same. Interesting. There is clearly a big.LITLLE penalty being paid in this benchmark, but this has been highlighted before on other workloads.
 

Top Liked Posts

Our Apps
Get our official app!
The best way to access XDA on your phone
Nav Gestures
Add swipe gestures to any Android
One Handed Mode
Eases uses one hand with your phone